具有不同值的查询执行不同

Question

让我们用测试数据创建2个表：

1)

create table test_users(
   id serial primary key
);

insert into test_users
select * from generate_series(1,750000);

2)

create table test_posts(
  id bigserial primary key,
  user_id INT
);

CREATE index idx ON test_posts(user_id, id);

INSERT INTO test_posts (user_id) SELECT 2 FROM generate_series (1,30038);
INSERT INTO test_posts (user_id) SELECT 3 FROM generate_series (1,31036);
INSERT INTO test_posts (user_id) SELECT 4 FROM generate_series (1,24032);
INSERT INTO test_posts (user_id) SELECT 5 FROM generate_series (1,93094);
INSERT INTO test_posts (user_id) SELECT 6 FROM generate_series (1,31033);
INSERT INTO test_posts (user_id) SELECT 7 FROM generate_series (1,62063);
INSERT INTO test_posts (user_id) SELECT 724309 FROM generate_series (1,2);
INSERT INTO test_posts (user_id) SELECT 724306 FROM generate_series (1,1);

现在，让我们检查行的数量，例如从id表中发布每个用户的最大test_posts：

SELECT user_id, count(*) , max(id) as maxid
FROM test_posts 
GROUP BY user_id
order by count(*)

所以，user_id = 724306只有一个帖子和（在我的情况下）post maxid是：271299。

user_id = 5在表中有大多数帖子，在我的情况下maxid是：178200

现在我运行这个查询：

SELECT test_posts.id
FROM test_posts 
JOIN test_users
ON test_posts.user_id = test_users.id 
WHERE test_posts.id = 271299  

explain analyze显示：

"Merge Join  (cost=8.88..9.06 rows=1 width=8) (actual time=146.561..146.563 rows=1 loops=1)"
"  Merge Cond: (test_users.id = test_posts.user_id)"
"  ->  Index Only Scan using test_users_pkey on test_users  (cost=0.42..22808.42 rows=750000 width=4) (actual time=0.012..110.091 rows=724307 loops=1)"
"        Heap Fetches: 724307"
"  ->  Sort  (cost=8.45..8.46 rows=1 width=12) (actual time=0.012..0.013 rows=1 loops=1)"
"        Sort Key: test_posts.user_id"
"        Sort Method: quicksort  Memory: 25kB"
"        ->  Index Scan using test_posts_pkey on test_posts  (cost=0.42..8.44 rows=1 width=12) (actual time=0.007..0.007 rows=1 loops=1)"
"              Index Cond: (id = 271299)"
"Planning time: 0.214 ms"
"Execution time: 146.588 ms"

现在运行相同的查询，只需使用178200：

SELECT test_posts.id
FROM test_posts 
JOIN test_users
ON test_posts.user_id = test_users.id 
WHERE test_posts.id = 178200  

explain analyze来自：

"Merge Join  (cost=8.88..9.06 rows=1 width=8) (actual time=0.040..0.042 rows=1 loops=1)"
"  Merge Cond: (test_users.id = test_posts.user_id)"
"  ->  Index Only Scan using test_users_pkey on test_users  (cost=0.42..22808.42 rows=750000 width=4) (actual time=0.016..0.018 rows=6 loops=1)"
"        Heap Fetches: 6"
"  ->  Sort  (cost=8.45..8.46 rows=1 width=12) (actual time=0.017..0.018 rows=1 loops=1)"
"        Sort Key: test_posts.user_id"
"        Sort Method: quicksort  Memory: 25kB"
"        ->  Index Scan using test_posts_pkey on test_posts  (cost=0.42..8.44 rows=1 width=12) (actual time=0.011..0.012 rows=1 loops=1)"
"              Index Cond: (id = 178200)"
"Planning time: 0.249 ms"
"Execution time: 0.072 ms"

导致执行时间差异的原因是什么？

Answer 1

确认了！（pg-9.4.5这里）

下面的技巧将使第一个查询再次快速。

---

EXPLAIN (ANALYZE, BUFFERS)
SELECT p.id
FROM test_posts p
LEFT JOIN test_users u
ON p.user_id = u.id AND u.id IS NOT NULL
WHERE p.id = 271299
        ;

---

也确认在10.3上。